POLO - Point-based, multi-class animal detection

2024-10-15

arXiv.org

AI Summary

In wildlife drone monitoring, multi-class animal detection typically relies on costly, labor-intensive bounding-box annotations with inherent label ambiguity. To address this, we propose the first end-to-end multi-class animal detection framework supervised solely by point-level labels. Building upon YOLOv8, we redesign the prediction head and post-processing pipeline, introducing a point-level IoU loss, center-point refinement strategy, and density-aware non-maximum suppression (NMS) to enable precise detection and counting driven exclusively by single-point localization. Evaluated on an aerial waterbird dataset, our method achieves significantly higher counting accuracy than standard YOLOv8 under identical annotation cost and demonstrates strong robustness in high-density scenes involving thousands of individuals. This work establishes a novel paradigm for multi-class fine-grained animal detection under point supervision.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Object Detection & Categorization
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Security and Privacy: Browser and end-point security
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Automated wildlife surveys based on drone imagery and object detection technology are a powerful and increasingly popular tool in conservation biology. Most detectors require training images with annotated bounding boxes, which are tedious, expensive, and not always unambiguous to create. To reduce the annotation load associated with this practice, we develop POLO, a multi-class object detection model that can be trained entirely on point labels. POLO is based on simple, yet effective modifications to the YOLOv8 architecture, including alterations to the prediction process, training losses, and post-processing. We test POLO on drone recordings of waterfowl containing up to multiple thousands of individual birds in one image and compare it to a regular YOLOv8. Our experiments show that at the same annotation cost, POLO achieves improved accuracy in counting animals in aerial imagery.

Problem

• Reduces annotation load for wildlife detection models
• Enables training on point labels instead of bounding boxes
• Improves accuracy in counting animals in aerial imagery

Innovation

• Uses point labels for multi-class detection
• Modifies YOLOv8 architecture for efficiency
• Improves accuracy in aerial animal counting